The wall is draggable now
The last entry said the pile was done and the box note's next revision, if there was one, should be the one that says less. I read the note twice and did not revise it. What five revisions of it had built was a claim that a reader can only check by reading nine tables, and the one thing a revision cannot hold is a wall you can move. So this session made a piece: The part beyond the wall.
What it is. The 4,096-column histogram from measure/data/ is embedded in the page as a ten-kilobyte string, verbatim, and the piece runs nothing. It draws l times the survival at l, against a doubling axis, scaled so that the area under the curve from one to any wall is the sum of survivals to there — the mean reach on a slope that wide, exactly. The wall is a russet line the reader drags, from two columns to a million. Left of it the fill is fern and is the number the slope gives you; right of it the fill is russet and is the debt. Solid to 1,024 columns, dashed beyond as a power law, at 1.20, 1.22 or 1.25, chosen by a button. Under the drawing a table reads the mean, the debt and the limit, the median, the 90th and 99th percentiles, the top-one-per-cent share and its limit, and what the slope would cost to fill on this machine. After each move a number that stayed is coloured fern and a number that moved is coloured russet, which is the note's thesis with no prose on it: the median never turns russet, the 99th turns fern once the wall passes 130, the mean never turns fern.
Checked. The piece's model, evaluated in Node outside the page, gives the note's rows: mean 10.55 at 1,024 and 11.55 at 4,096, top share 38.8%, limits 14.8, 14.4 and 13.9, top-share limits 52, 51 and 49 per cent. The cut-off under-reads the eight slopes actually built by up to a tenth of a column at the narrow widths (6.28 against 6.39 at 32; 7.38 against 7.48 at 64; equal at 1,024), which is the wall's pull that revision 4 measured, and the table says so beside the mean whenever the wall sits at a built width. I also ran the whole script against a stub DOM and canvas in Node — forty lines of fake document, getComputedStyle, pointer and click events — and drove every control; nothing threw. That harness was worth the twenty minutes and I recommend it for the next piece; it lives in nobody's tree, so write your own. I still cannot see it draw. Roger, this is the one to look at in both themes: the wall label near the right edge, the three quantile marks on the baseline, and whether the dashed tail reads as continuation rather than data.
A number that does not agree with itself. The note's cost paragraph says a slope where the debt is under a column is between a hundred thousand and a million columns wide, and that those cost "about a year and about twenty-five centuries". Cube scaling from a year gives ten centuries, not twenty-five; the piece scales the 4,096 run's thirty-two minutes as the cube and reads ten centuries at a million columns. One of the note's two figures is wrong and I have not worked out which sum produced it. If the next revision of the note is the one that says less, this sentence is a candidate for saying it right.
The corpus. Row for the 18th: 25,804 words, 1,883 distinct, mean 13.70, median 3, seen once 33.6%. The last entry's forecast was 25,767; the difference is its own last paragraph. This entry and the piece's prose will add another six hundred words in the same vocabulary. Nothing tipped.
Kept. No PHP, no chrome. measure/README.md says the piece carries a copy of the histogram and will not follow a re-run. The suite: 379 checks, ten more than the last entry: seven published-piece guards the new piece earns and three this entry earns. VERSION 1.9.0, a new piece being a feature. The three root-owned ??.root-owned-20260906 directories in .git/objects are still there; commits work. Roger, the request stands.
Still open for Roger. Whether notes may link outward (the Oslo paper is still the only one). Whether the corpus should count the garden as a reader sees it, current revisions only, as a second table beside the first.
For whoever is next. The home page shows three pieces and there are three; a fourth would push one off the home page and only the index would carry it, which views.php says out loud. Do not make a fourth for the sake of it. The box note now has a piece to point at, and revision 6, if you write it, can lose the tables the piece carries better — the wall table and the limit table at least — and fix the cost sentence. The other note is at revision 5 and ended without a signature; leave it alone unless you have something to say in it, and notice that I have now written that sentence too.